JungMAE (pretrained encoder, GeoNIRS release)

Weights of the masked-autoencoder fNIRS foundation model of Jung, Lee and An (2025), as implemented and trained by the GeoNIRS authors (Li et al., 2026), for braindecode.models.JungMAE. They are the encoder of checkpoint/jung_mae.pt of GeoNIRS/GeoNIRS at commit 81f8995, converted by convert.py.

from braindecode.models import JungMAE

model = JungMAE.from_pretrained(
    "braindecode/jungmae-pretrained", n_chans=40, n_outputs=2, n_times=60
)
features = model(x, return_features=True)["features"]

The classification head (final_layer) is not part of the release: it reads n_chans * n_patches * embed_dim features, so from_pretrained builds it for your montage and n_outputs. Fine-tune or linear-probe before use.

Intended input

The inputs of the GeoNIRS paper (Appendix B.1): HbO and HbR in µmol/L resampled to 5 Hz, each chromophore z-scored per recording over all its valid channels and time points.

Changes from the original release

convert.py unwraps the {"model", "model_config", ...} payload (its configuration equals the class defaults), drops the pretraining decoder (input projection, mask token, layers, prediction layer) and writes the encoder with save_pretrained. No value is changed: the encoder gives the features of the original file bit for bit on CPU in float32.

License and citation

The source repository has no licence file; the weights are by the GeoNIRS authors. Please cite the method and the release:

@inproceedings{jung2025mae,
  author    = {Jung, Euijin and Lee, Hyunmin and An, Jinung},
  title     = {{fNIRS} Foundation Model for Few-Shot Based {fNIRS} Classification},
  booktitle = {2025 13th International Conference on Brain-Computer Interface (BCI)},
  year      = {2025},
  doi       = {10.1109/BCI65088.2025.10931275},
}

@article{li2026geonirs,
  author  = {Li, Zongsheng and Zhu, Shengyuan and Zhang, Zimo and Li, Hanyao and Wang, Shengming and Liu, Quanying},
  title   = {{GeoNIRS}: Geometry-Aware Pretraining for Heterogeneous {fNIRS} Representation Learning},
  journal = {arXiv preprint arXiv:2610.17845},
  year    = {2026},
}
Downloads last month
-
Safetensors
Model size
1.78M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Paper for braindecode/jungmae-pretrained